linux_dsm_epyc7002/drivers/gpu/drm/i915/intel_hangcheck.c

487 lines
14 KiB
C
Raw Normal View History

/*
* Copyright © 2016 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*
*/
#include "i915_drv.h"
static bool
ipehr_is_semaphore_wait(struct intel_engine_cs *engine, u32 ipehr)
{
if (INTEL_GEN(engine->i915) >= 8) {
return (ipehr >> 23) == 0x1c;
} else {
ipehr &= ~MI_SEMAPHORE_SYNC_MASK;
return ipehr == (MI_SEMAPHORE_MBOX | MI_SEMAPHORE_COMPARE |
MI_SEMAPHORE_REGISTER);
}
}
static struct intel_engine_cs *
semaphore_wait_to_signaller_ring(struct intel_engine_cs *engine, u32 ipehr,
u64 offset)
{
struct drm_i915_private *dev_priv = engine->i915;
struct intel_engine_cs *signaller;
enum intel_engine_id id;
if (INTEL_GEN(dev_priv) >= 8) {
for_each_engine(signaller, dev_priv, id) {
if (engine == signaller)
continue;
if (offset == signaller->semaphore.signal_ggtt[engine->hw_id])
return signaller;
}
} else {
u32 sync_bits = ipehr & MI_SEMAPHORE_SYNC_MASK;
for_each_engine(signaller, dev_priv, id) {
if(engine == signaller)
continue;
if (sync_bits == signaller->semaphore.mbox.wait[engine->hw_id])
return signaller;
}
}
DRM_DEBUG_DRIVER("No signaller ring found for %s, ipehr 0x%08x, offset 0x%016llx\n",
engine->name, ipehr, offset);
return ERR_PTR(-ENODEV);
}
static struct intel_engine_cs *
semaphore_waits_for(struct intel_engine_cs *engine, u32 *seqno)
{
struct drm_i915_private *dev_priv = engine->i915;
void __iomem *vaddr;
u32 cmd, ipehr, head;
u64 offset = 0;
int i, backwards;
/*
* This function does not support execlist mode - any attempt to
* proceed further into this function will result in a kernel panic
* when dereferencing ring->buffer, which is not set up in execlist
* mode.
*
* The correct way of doing it would be to derive the currently
* executing ring buffer from the current context, which is derived
* from the currently running request. Unfortunately, to get the
* current request we would have to grab the struct_mutex before doing
* anything else, which would be ill-advised since some other thread
* might have grabbed it already and managed to hang itself, causing
* the hang checker to deadlock.
*
* Therefore, this function does not support execlist mode in its
* current form. Just return NULL and move on.
*/
if (engine->buffer == NULL)
return NULL;
ipehr = I915_READ(RING_IPEHR(engine->mmio_base));
if (!ipehr_is_semaphore_wait(engine, ipehr))
return NULL;
/*
* HEAD is likely pointing to the dword after the actual command,
* so scan backwards until we find the MBOX. But limit it to just 3
* or 4 dwords depending on the semaphore wait command size.
* Note that we don't care about ACTHD here since that might
* point at at batch, and semaphores are always emitted into the
* ringbuffer itself.
*/
head = I915_READ_HEAD(engine) & HEAD_ADDR;
backwards = (INTEL_GEN(dev_priv) >= 8) ? 5 : 4;
vaddr = (void __iomem *)engine->buffer->vaddr;
for (i = backwards; i; --i) {
/*
* Be paranoid and presume the hw has gone off into the wild -
* our ring is smaller than what the hardware (and hence
* HEAD_ADDR) allows. Also handles wrap-around.
*/
head &= engine->buffer->size - 1;
/* This here seems to blow up */
cmd = ioread32(vaddr + head);
if (cmd == ipehr)
break;
head -= 4;
}
if (!i)
return NULL;
*seqno = ioread32(vaddr + head + 4) + 1;
if (INTEL_GEN(dev_priv) >= 8) {
offset = ioread32(vaddr + head + 12);
offset <<= 32;
offset |= ioread32(vaddr + head + 8);
}
return semaphore_wait_to_signaller_ring(engine, ipehr, offset);
}
static int semaphore_passed(struct intel_engine_cs *engine)
{
struct drm_i915_private *dev_priv = engine->i915;
struct intel_engine_cs *signaller;
u32 seqno;
engine->hangcheck.deadlock++;
signaller = semaphore_waits_for(engine, &seqno);
if (signaller == NULL)
return -1;
if (IS_ERR(signaller))
return 0;
/* Prevent pathological recursion due to driver bugs */
if (signaller->hangcheck.deadlock >= I915_NUM_ENGINES)
return -1;
if (i915_seqno_passed(intel_engine_get_seqno(signaller), seqno))
return 1;
/* cursory check for an unkickable deadlock */
if (I915_READ_CTL(signaller) & RING_WAIT_SEMAPHORE &&
semaphore_passed(signaller) < 0)
return -1;
return 0;
}
static void semaphore_clear_deadlocks(struct drm_i915_private *dev_priv)
{
struct intel_engine_cs *engine;
enum intel_engine_id id;
for_each_engine(engine, dev_priv, id)
engine->hangcheck.deadlock = 0;
}
static bool instdone_unchanged(u32 current_instdone, u32 *old_instdone)
{
u32 tmp = current_instdone | *old_instdone;
bool unchanged;
unchanged = tmp == *old_instdone;
*old_instdone |= tmp;
return unchanged;
}
static bool subunits_stuck(struct intel_engine_cs *engine)
{
struct drm_i915_private *dev_priv = engine->i915;
struct intel_instdone instdone;
struct intel_instdone *accu_instdone = &engine->hangcheck.instdone;
bool stuck;
int slice;
int subslice;
if (engine->id != RCS)
return true;
intel_engine_get_instdone(engine, &instdone);
/* There might be unstable subunit states even when
* actual head is not moving. Filter out the unstable ones by
* accumulating the undone -> done transitions and only
* consider those as progress.
*/
stuck = instdone_unchanged(instdone.instdone,
&accu_instdone->instdone);
stuck &= instdone_unchanged(instdone.slice_common,
&accu_instdone->slice_common);
for_each_instdone_slice_subslice(dev_priv, slice, subslice) {
stuck &= instdone_unchanged(instdone.sampler[slice][subslice],
&accu_instdone->sampler[slice][subslice]);
stuck &= instdone_unchanged(instdone.row[slice][subslice],
&accu_instdone->row[slice][subslice]);
}
return stuck;
}
static enum intel_engine_hangcheck_action
head_stuck(struct intel_engine_cs *engine, u64 acthd)
{
if (acthd != engine->hangcheck.acthd) {
/* Clear subunit states on head movement */
memset(&engine->hangcheck.instdone, 0,
sizeof(engine->hangcheck.instdone));
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_ACTIVE_HEAD;
}
if (!subunits_stuck(engine))
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_ACTIVE_SUBUNITS;
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_DEAD;
}
static enum intel_engine_hangcheck_action
engine_stuck(struct intel_engine_cs *engine, u64 acthd)
{
struct drm_i915_private *dev_priv = engine->i915;
enum intel_engine_hangcheck_action ha;
u32 tmp;
ha = head_stuck(engine, acthd);
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
if (ha != ENGINE_DEAD)
return ha;
if (IS_GEN2(dev_priv))
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_DEAD;
/* Is the chip hanging on a WAIT_FOR_EVENT?
* If so we can simply poke the RB_WAIT bit
* and break the hang. This should work on
* all but the second generation chipsets.
*/
tmp = I915_READ_CTL(engine);
if (tmp & RING_WAIT) {
i915_handle_error(dev_priv, 0,
"Kicking stuck wait on %s",
engine->name);
I915_WRITE_CTL(engine, tmp);
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_WAIT_KICK;
}
if (INTEL_GEN(dev_priv) >= 6 && tmp & RING_WAIT_SEMAPHORE) {
switch (semaphore_passed(engine)) {
default:
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_DEAD;
case 1:
i915_handle_error(dev_priv, 0,
"Kicking stuck semaphore on %s",
engine->name);
I915_WRITE_CTL(engine, tmp);
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_WAIT_KICK;
case 0:
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_WAIT;
}
}
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_DEAD;
}
static void hangcheck_load_sample(struct intel_engine_cs *engine,
struct intel_engine_hangcheck *hc)
{
/* We don't strictly need an irq-barrier here, as we are not
* serving an interrupt request, be paranoid in case the
* barrier has side-effects (such as preventing a broken
* cacheline snoop) and so be sure that we can see the seqno
* advance. If the seqno should stick, due to a stale
* cacheline, we would erroneously declare the GPU hung.
*/
if (engine->irq_seqno_barrier)
engine->irq_seqno_barrier(engine);
hc->acthd = intel_engine_get_active_head(engine);
hc->seqno = intel_engine_get_seqno(engine);
}
static void hangcheck_store_sample(struct intel_engine_cs *engine,
const struct intel_engine_hangcheck *hc)
{
engine->hangcheck.acthd = hc->acthd;
engine->hangcheck.seqno = hc->seqno;
engine->hangcheck.action = hc->action;
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
engine->hangcheck.stalled = hc->stalled;
}
static enum intel_engine_hangcheck_action
hangcheck_get_action(struct intel_engine_cs *engine,
const struct intel_engine_hangcheck *hc)
{
if (engine->hangcheck.seqno != hc->seqno)
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_ACTIVE_SEQNO;
if (i915_seqno_passed(hc->seqno, intel_engine_last_submit(engine)))
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
return ENGINE_IDLE;
return engine_stuck(engine, hc->acthd);
}
static void hangcheck_accumulate_sample(struct intel_engine_cs *engine,
struct intel_engine_hangcheck *hc)
{
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
unsigned long timeout = I915_ENGINE_DEAD_TIMEOUT;
hc->action = hangcheck_get_action(engine, hc);
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
/* We always increment the progress
* if the engine is busy and still processing
* the same request, so that no single request
* can run indefinitely (such as a chain of
* batches). The only time we do not increment
* the hangcheck score on this ring, if this
* engine is in a legitimate wait for another
* engine. In that case the waiting engine is a
* victim and we want to be sure we catch the
* right culprit. Then every time we do kick
* the ring, make it as a progress as the seqno
* advancement might ensure and if not, it
* will catch the hanging engine.
*/
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
switch (hc->action) {
case ENGINE_IDLE:
case ENGINE_ACTIVE_SEQNO:
/* Clear head and subunit states on seqno movement */
hc->acthd = 0;
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
memset(&engine->hangcheck.instdone, 0,
sizeof(engine->hangcheck.instdone));
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
/* Intentional fall through */
case ENGINE_WAIT_KICK:
case ENGINE_WAIT:
engine->hangcheck.action_timestamp = jiffies;
break;
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
case ENGINE_ACTIVE_HEAD:
case ENGINE_ACTIVE_SUBUNITS:
/* Seqno stuck with still active engine gets leeway,
* in hopes that it is just a long shader.
*/
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
timeout = I915_SEQNO_DEAD_TIMEOUT;
break;
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
case ENGINE_DEAD:
break;
default:
MISSING_CASE(hc->action);
}
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
hc->stalled = time_after(jiffies,
engine->hangcheck.action_timestamp + timeout);
}
static void hangcheck_declare_hang(struct drm_i915_private *i915,
unsigned int hung,
unsigned int stuck)
{
struct intel_engine_cs *engine;
char msg[80];
unsigned int tmp;
int len;
/* If some rings hung but others were still busy, only
* blame the hanging rings in the synopsis.
*/
if (stuck != hung)
hung &= ~stuck;
len = scnprintf(msg, sizeof(msg),
"%s on ", stuck == hung ? "No progress" : "Hang");
for_each_engine_masked(engine, i915, hung, tmp)
len += scnprintf(msg + len, sizeof(msg) - len,
"%s, ", engine->name);
msg[len-2] = '\0';
return i915_handle_error(i915, hung, "%s", msg);
}
/*
* This is called when the chip hasn't reported back with completed
* batchbuffers in a long time. We keep track per ring seqno progress and
* if there are no progress, hangcheck score for that ring is increased.
* Further, acthd is inspected to see if the ring is stuck. On stuck case
* we kick the ring. If we see no progress on three subsequent calls
* we assume chip is wedged and try to fix it by resetting the chip.
*/
static void i915_hangcheck_elapsed(struct work_struct *work)
{
struct drm_i915_private *dev_priv =
container_of(work, typeof(*dev_priv),
gpu_error.hangcheck_work.work);
struct intel_engine_cs *engine;
enum intel_engine_id id;
unsigned int hung = 0, stuck = 0;
int busy_count = 0;
if (!i915.enable_hangcheck)
return;
if (!READ_ONCE(dev_priv->gt.awake))
return;
if (i915_terminally_wedged(&dev_priv->gpu_error))
return;
/* As enabling the GPU requires fairly extensive mmio access,
* periodically arm the mmio checker to see if we are triggering
* any invalid access.
*/
intel_uncore_arm_unclaimed_mmio_detection(dev_priv);
for_each_engine(engine, dev_priv, id) {
struct intel_engine_hangcheck cur_state, *hc = &cur_state;
const bool busy = intel_engine_has_waiter(engine);
semaphore_clear_deadlocks(dev_priv);
hangcheck_load_sample(engine, hc);
hangcheck_accumulate_sample(engine, hc);
hangcheck_store_sample(engine, hc);
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
if (engine->hangcheck.stalled) {
hung |= intel_engine_flag(engine);
drm/i915: Decouple hang detection from hangcheck period Hangcheck state accumulation has gained more steps along the years, like head movement and more recently the subunit inactivity check. As the subunit sampling is only done if the previous state check showed inactivity, we have added more stages (and time) to reach a hang verdict. Asymmetric engine states led to different actual weight of 'one hangcheck unit' and it was demonstrated in some hangs that due to difference in stages, simpler engines were accused falsely of a hang as their scoring was much more quicker to accumulate above the hang treshold. To completely decouple the hangcheck guilty score from the hangcheck period, convert hangcheck score to a rough period of inactivity measurement. As these are tracked as jiffies, they are meaningful also across reset boundaries. This makes finding a guilty engine more accurate across multi engine activity scenarios, especially across asymmetric engines. We lose the ability to detect cross batch malicious attempts to hinder the progress. Plan is to move this functionality to be part of context banning which is more natural fit, later in the series. v2: use time_before macros (Chris) reinstate the pardoning of moving engine after hc (Chris) v3: avoid global state for per engine stall detection (Chris) v4: take timeline last retirement into account (Chris) v5: do debug print on pardoning, split out retirement timestamp (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
2016-11-18 20:09:04 +07:00
if (hc->action != ENGINE_DEAD)
stuck |= intel_engine_flag(engine);
}
busy_count += busy;
}
if (hung)
hangcheck_declare_hang(dev_priv, hung, stuck);
/* Reset timer in case GPU hangs without another request being added */
if (busy_count)
i915_queue_hangcheck(dev_priv);
}
void intel_engine_init_hangcheck(struct intel_engine_cs *engine)
{
memset(&engine->hangcheck, 0, sizeof(engine->hangcheck));
}
void intel_hangcheck_init(struct drm_i915_private *i915)
{
INIT_DELAYED_WORK(&i915->gpu_error.hangcheck_work,
i915_hangcheck_elapsed);
}
#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
#include "selftests/intel_hangcheck.c"
#endif